1
UNVEILING GENERATIVE AI: A NEW FRONTIER

Okay, let's go back to basics. What is generative AI (artificial intelligence)? How does it work? And what is the technology capable of? You'll find out in this chapter as we delve under the hood of GenAI, tug at a few wires, and examine what the heck's going on in there.

I really want this chapter to give you an inspiring feel for the many things that GenAI is capable of – but also to drive home the point that GenAI isn't just about ChatGPT (Chat Generative Pre-trained Transformer). Sure, ChatGPT is a prime example of GenAI (and it certainly hoovers up the majority of GenAI's press), but there's a lot more to GenAI than ChatGPT … as you'll find out in this chapter.

What Is Generative AI? A Quick Explanation

Actually, let's start by defining artificial intelligence (AI) in its broadest sense. The term “AI” refers to computer algorithms that can effectively simulate human cognitive processes, like learning, decision-making, and problem-solving.

GenAI is a groundbreaking subset of AI – the cutting edge of the cutting edge – that is able to create new content based on patterns and structures it has learned from existing data. Like any AI, GenAI tools are given enormous amounts of data to learn from (what's known as “training data”). They learn from the training data, and then use the patterns or rules that they've learned to create new content that's similar to, but not exactly the same as, the data they have been trained on.

An example or two

A good example is DALL-E 2 – the text-to-art platform that allows anyone to generate artworks. Or, of course, there's ChatGPT, the language model that can create text based on conversational text prompts. (You'll find a much bigger list of GenAI tools in the Appendix, by the way.) These systems learn from huge training datasets – ChatGPT, for example, was trained on vast amounts of text from the internet, including web pages, articles, and books.

Text and images are perhaps two of the best-known uses of GenAI so far, but the technology is capable of so much more. With GenAI you can generate product designs, computer code, music, video, voices, and even entire visual worlds. We'll talk more about GenAI's capabilities later in the chapter, but to whet your appetite, imagine being able to create unique video game worlds rendered in real time, or have a book written just for you, or have your favorite celebrity read you today's news. The possibilities are mind-blowing.

The capabilities are already quite impressive, but in the future, GenAI systems will be able to create pretty much anything that humans can. And this, in turn, means GenAI can turn anyone and everyone into an author, musician, computer programmer, filmmaker, or other type of creator.

How GenAI differs from the AI we are used to

AI is increasingly part of the world around us, including the search results you get on your phone, your conversations with Alexa, and the movie recommendations that Netflix serves up on an evening. So what makes GenAI different to these “traditional” AI tools? (I realize it sounds strange to refer to “traditional AI” when it's hardly been around that long, but I do so to distinguish between the AI that we're already used to in everyday life and this new evolution of AI systems. One technical term for traditional AI is “discriminative AI.”)

Traditional AI systems also learn from large amounts of data, but they deliver a different output. Traditional AI systems are used to make predictions based on existing data. And we use those predictions to help us make better decisions, at work and in everyday life. This could span anything from listening to new music on Spotify, and viewing recommended products on Amazon, to identifying which of your company's customers are most likely to buy a certain product.

This new wave of GenAI tools goes even further, by creating new content based on existing data. In other words, GenAI isn't just about simulating human cognitive processes like decision-making and problem-solving – it's about simulating human creativity.

To further illustrate the difference, imagine you're playing computer chess. The computer knows all the rules, can predict your moves, and makes its own moves. It's not inventing new chess moves; rather, it's selecting the right move based on existing strategies. That's traditional AI – it's like a master strategist who can make smart decisions within a set of rules. And it does it very well. But GenAI? Well, that could, in theory, come up with new ways of playing chess that we haven't invented yet.

So, traditional AI excels at pattern recognition, while GenAI excels at pattern creation. Which is very cool indeed.

That said, GenAI and traditional AI aren't mutually exclusive. They could be used in tandem to provide even more powerful solutions. For example, a traditional AI could analyze user behavior from your company's website, and then a GenAI could use the analysis to create personalized content for users.

How Does Generative AI Work?

Think of it like learning to draw by looking at many pictures. After looking at many pictures, you try to draw something new on your own. GenAI does something similar: it “looks” at a lot of data (text, visual, or other), learns the patterns within that data, and then tries to create something new that fits those patterns.

So, in very simple terms, GenAI is like an artist or writer that has studied lots of existing works and then tries to create its own work based on what it has learned. This process is powered by complex algorithms that mimic how our brain works in order to learn from data and identify patterns.

That's the super-simple explanation. Let's get a little more technical.

Introducing machine learning and neural networks

We'll talk more about the evolution of GenAI in Chapter 2, but as a quick primer, GenAI grew out of a field of AI study and practice called “machine learning” – indeed all of the AI we see today is based on machine learning. While traditional computer algorithms are coded by a human to tell a machine exactly how to do a particular job, machine-learning algorithms are able to make decisions based on what they learn from the data. The more data they're fed, the better they get at this process.

Another term you'll need to get to grips with is “neural networks,” as this is the core technology that GenAI is built upon. A neural network is essentially an advanced machine-learning model inspired by the way human brains work. While a less complex machine-learning model may need some human intervention in the process, a neural network has the ability to learn and make decisions by itself, and can even learn from its own errors – rather like the way a human learns through a process of trial and error.

Here's how a neural network works:

• Training: neural networks learn through a process called “training.” During training, the neural network is fed a lot of data (which could be text, images, or whatever), and it learns to identify patterns and relationships in the data.
• Learning: as the neural network is exposed to more and more data – and we really are talking about vast amounts of data – it gradually gets better at identifying patterns and understanding the underlying rules that govern the data.
• Layers: neural networks are organized into layers, and each layer is responsible for identifying different types of patterns. The initial layers might identify simple patterns, and as we move deeper, the layers are able to identify more complex patterns.
• Generative models: GenAI often uses specific types of neural networks called “generative models.” In addition to recognizing patterns, generative models are able to generate new data that is similar to (but not exactly the same as) the data they were trained on.
• Input and output: once the generative model is trained, you can give it an input (e.g., a partial image or a text prompt), and it will generate an output (like a completed image or a piece of text) based on what it learned during training.
• Randomness: GenAI introduces a certain amount of randomness in the generation process, which means that it can produce slightly different outputs each time, even when given the same input over and over again.

Examples of generative models

I just mentioned generative models, the neural networks that enable GenAI to create new content. Here are some examples of generative models used in GenAI applications:

• Large language models (LLMs): by gobbling up large amounts of text, LLMs learn the semantic relationships between words and use that data to generate more language. An example of an LLM is GPT-4, created by OpenAI, which powers the ChatGPT tool.
• Generative adversarial networks (GANs): these work by pitting two competing algorithms against each other, one tasked with generating data that resembles its training data and the other tasked with trying to tell whether the output is real or generated. This type of model is typically used to create images, sounds, or even video (including deepfakes).
• Variational autoencoders: this is a...